The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates a workflow for ingesting and analyzing external text (such as public advisories or vendor reports) fetched from the web. This creates a surface for indirect prompt injection where malicious instructions embedded in the processed external pages could influence the agent's summary or report output.
Ingestion points: External content from URLs fetched and converted via the defuddle.md service (SKILL.md).
Boundary markers: None identified; the instructions do not suggest wrapping external content in delimiters or using instructions to ignore embedded commands.
Capability inventory: Summarization, report generation, and security assessment across all files.
Sanitization: No specific sanitization, filtering, or validation of the fetched external content is described.
[DATA_EXFILTRATION]: The skill includes instructions to use a third-party service (defuddle.md) for processing URLs. This constitutes a network operation to a non-whitelisted domain. The author provides a security warning noting that target URLs are sent to this third party, which could leak information about client-owned hosts or engagement targets.
[EXTERNAL_DOWNLOADS]: The skill suggests fetching and converting remote web content using curl from a non-standard third-party domain to support its analysis workflow.
[COMMAND_EXECUTION]: The skill contains instructional blocks that prompt the execution of shell commands (curl) to interact with external services and retrieve data.

reporting-security-findings